CN108090961A - A kind of quick error compensation method in three-dimensional laser point cloud imaging - Google Patents
A kind of quick error compensation method in three-dimensional laser point cloud imaging Download PDFInfo
- Publication number
- CN108090961A CN108090961A CN201810021605.7A CN201810021605A CN108090961A CN 108090961 A CN108090961 A CN 108090961A CN 201810021605 A CN201810021605 A CN 201810021605A CN 108090961 A CN108090961 A CN 108090961A
- Authority
- CN
- China
- Prior art keywords
- point
- key frame
- characteristic
- point cloud
- adjustment
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/10—Constructive solid geometry [CSG] using solid primitives, e.g. cylinders, cubes
Landscapes
- Physics & Mathematics (AREA)
- Geometry (AREA)
- Engineering & Computer Science (AREA)
- Computer Graphics (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Optical Radar Systems And Details Thereof (AREA)
- Length Measuring Devices By Optical Means (AREA)
Abstract
A kind of quick error compensation method in being imaged the invention discloses three-dimensional laser point cloud, including：Key frame extraction, key frame registration, adjustment detection, adjustment optimization.Key frame extraction is responsible for selecting the point cloud used in subsequent registration from whole cloud datas that laser radar obtains；Key frame registration link reduces image error by the registration between adjacent two key frame；Adjustment detection be responsible for judging in key frame registration link whether appearance point cloud matching error；Adjustment optimization link is optimized for the point cloud matching mistake occurred.The present invention provides a kind of method of quick adjustment for mobile laser radar real time imagery, and algorithm zmodem can be adapted for the three-dimensional point cloud imaging of a variety of varying environments.
Description
Technical field
A kind of quick error compensation method in being imaged the present invention relates to three-dimensional laser point cloud, belongs to remote sensing and mapping, computer
Vision and mode identification technology, the three dimensional point cloud obtained suitable for any 3D digitizer.
Background technology
The adjustment of three-dimensional laser point cloud imaging is always that computer vision is extremely closed with pattern-recognition, remote sensing and survey field
The problem of note, lasting cumulative data will necessarily cause track drift error, it is necessary to optimize adjustment.Previous adjustment Algorithm is most
Global optimization adjustment is carried out afterwards, and such speed is slow, without real-time, is unable to reach real time imagery.
The content of the invention
The technical problem to be solved in the present invention is：In mobile laser radar map building imaging, laser obtains newly with it
One frame point cloud is obtained in a manner of cumulative, and cumulative errors can be caused by persistently adding up, and the present invention provides a kind of three-dimensional laser point
Quick error compensation method in cloud imaging.The distribution characteristics of this method from cloud is started with, and is extracted representational characteristic point, is used spy
Sign point is rapidly performed by registration optimization error while imaging, substantially reduces cumulative errors, and algorithm zmodem is suitble to appoint
What 3D digitizer obtains the search matching of three-dimensional point cloud.
The present invention solve the technical solution that uses of above-mentioned technical problem for：A kind of quick flat in three-dimensional laser point cloud imaging
Difference method, this method comprises the following steps：
Step (1) determines a fixed time interval, is chosen and closed in whole point clouds of acquisition based on the interval
Key frame carries out adjustment, reduces computational complexity；
In key frame points cloud, extract boundary point and planar point, method is step (2)：For each point, ask the point to
The mould of the vector sum of all the points in its certain distance, then divided by laser range finder to the distance of the point, obtain the flatness
I.e.Wherein i is the point calculated, and j is the consecutive points of i, and S is that the point of i both sides same number is formed
Point set, Pi、PjThe respectively coordinate of point i points j.Take n point of wherein c values maximum that it is minimum (close to take out c values as boundary point
0) n point is as planar point.
After step (3) determines characteristic point, feature association is carried out to each characteristic point.Feature association is exactly to seek this feature point
The distance of the linked character of the point into previous frame.
After step (4) feature association, each characteristic point is looked for there are one the characteristic distance d on previous frame data
T is converted to a posek=[tx,ty,tz,rx,ry,rz], it is all 0 to allow all characteristic distances as far as possible, this pose converts just
Movement of the laser radar between this two frame data, i.e. f (Pk,Tk)=dk→ 0, PkFor k moment characteristic point matrix, dkFor k when
Carve characteristic distance matrix, TkIt is converted for the pose at k-1 to k moment.Optimized using least square method and calculate Tk, according to TkUpdate point
Cloud position carries out feature association and updates distance d to calculate T againk, constantly Xun Huan is until d is less than given threshold or cycles secondary
Number reaches the upper limit；
Step (5) reduces if the average value of whole characteristic distances after optimization completion is more preceding than optimization, then it is assumed that adjustment is complete
Into；Conversely, then to carry out adjustment optimization：The point cloud of current key frame is mapped to after global coordinate system, it is possible to determine to swash
Position of the optical radar in the three-dimensional point cloud atlas created, in three-dimensional point cloud atlas, one cube is taken centered on this position
In vivo all point clouds are with allowing current key frame to be matched somebody with somebody again by step (2), step (3), step (4) with this partial dot cloud
Standard optimizes to complete adjustment, wherein cubical size is related with the measurement range of laser radar.
Wherein, registration error method for solving is as follows in the step (3)：
For step (3.1) if characteristic point is planar point, its linked character is a plane, it is necessary to be sought in previous frame data
Three points are looked for determine this plane, traversal finds point M nearest apart from characteristic point in previous keyframe first, then extracts and M
With scan line and the nearest point L of distance M, point N nearest with M in the adjacent scanning lines of M is extracted；
For step (3.2) if characteristic point is planar point, its linked character is a plane, it is necessary to be sought in previous frame data
Three points is looked for determine this plane：Point M nearest apart from characteristic point in traversal searching previous keyframe first, then M's are adjacent
The point N nearest with M in scan line.
The present invention compared with prior art the advantages of be：
(1) present invention carries out adjustment while imaging, substantially increases the efficiency being imaged in the actual environment.
(2) all there is similar spy as characteristic point in the boundary point in present invention extraction point cloud and planar point in various environment
Sign has universality, can supply to be applied to various different actual scenes.
(3) present invention can choose current optimum point pair automatically in feature association using ergodic algorithm in calculating process,
And then obtain optimal solution.This feature makes experimenter be improved work efficiency without manual selected point pair.
(4) present invention can detect discovery in time when certain adjustment fails, and compensate automatically, improve the Shandong of algorithm
Stick.
(5) the method for the invention, from environmental restrictions, can be lost inferior GPS indoors without using equipment such as GPS/IMU
Lock ring uses in border.
Description of the drawings
The method that Fig. 1 is the present invention realizes flow chart；
Fig. 2 is key frame extraction schematic diagram of the present invention；
Fig. 3 is plane characteristic point correlating method figure of the present invention；
Fig. 4 is edge feature point correlating method figure of the present invention；
Fig. 5 is three-dimensional imaging result point cloud chart of the present invention；
Fig. 6 is imaging results plane figure of the present invention；
Specific embodiment
Below in conjunction with the accompanying drawings and specific embodiment further illustrates the present invention.
As shown in Figure 1, the present invention includes following calculation procedure：Key frame extraction, key frame are registering, adjustment detects,
Adjustment optimizes.Key frame extraction is responsible for selecting the point used in subsequent registration from whole cloud datas that laser radar obtains
Cloud；Key frame registration link reduces image error by the registration between adjacent two key frame；Adjustment detection is responsible for sentencing
In disconnected key frame registration link whether appearance point cloud matching error；Adjustment optimization link is carried out for the point cloud matching mistake occurred
Optimization.It is as follows：
It now holds 16 line laser radars and obtains indoor point cloud progress map building, pass through key frame extraction, key frame here
4 steps such as registration, adjustment detection, adjustment optimization complete adjustment in real time while map building.
(1) key frame extraction
Determine a fixed time interval, choosing key frame in whole point clouds of acquisition based on the interval carries out
Adjustment reduces computational complexity.As shown in Fig. 2, setting one selects a frame data per X frames as key at intervals of X frame data
Frame carries out adjustment to each key frame.By taking n+1 parts as an example, during map building, the data meeting of key frame (n+1) X+1
Obtained pose transformed mappings are in global coordinate system in a manner of persistently adding up, by the data after the completion of mapping and upper one pass
Key frame nX+1 carries out registration again, and total pose conversion after optimizing mapping result and being optimized, this completes once put down
Difference.Wherein, laser radar is shorter in two crucial interframe movement distance S, and the feature repeated between two key frames is more, registration effect
Fruit is better.As the mobile distance S of formula (1) frame number X, laser radar movement velocity v and the frequency f included with each big window has
It closes, so smaller key frame interval should be set in the case where not influencing program operation speed and adjustment effects.In the example
Middle selection X is 10.
(2) key frame registration
After determining key frame, in key frame points cloud, boundary point and planar point are extracted.For each point, the point is asked to arrive
The mould of the vector sum of all the points in its certain distance, then divided by laser range finder to the distance of the point, it is smooth to obtain the point
Degree is
5 points of wherein c values maximum is taken to take out 5 points of c values minimum (close to 0) as planar point as boundary point.If
Point P is planar point, its linked character is a plane, it is necessary to find three points in previous frame data to determine that this is flat
Face.As shown in figure 3, M is the nearest points of distance P in previous frame data, L is to be with scan line and the nearest points of distance M, N with M
The point nearest with M in the adjacent scanning lines of M.If point P is boundary point, its linked character is straight line, it is necessary in previous frame
Two points are found in data to determine this straight line.As shown in figure 4, M is the nearest points of the distance P in of previous frame number 1, N is M's
The point nearest with M in adjacent scanning lines, wherein in three dimensions, the distance of point P to straight line MN equal to point P arrived straight line MN and
The distance of the plane vertical with plane PMN.Final two kinds of features are converted to the distance of certain plane, and formula (3) is plane equation.
Characteristic distance, wherein x can be obtained according to formula (4) after the completion of association0y0z0It is characterized point coordinates.
Ax+By+Cz+D=0 (3)
Each characteristic point finds a pose conversion T there are one the characteristic distance d on previous frame datak=[tx,
ty,tz,rx,ry,rz], it is all 0 to allow all characteristic distances as far as possible, and the conversion of this pose is exactly laser radar in this two frame data
Between movement, i.e. f (Pk,Tk)=dk→ 0, PkFor k moment characteristic point matrix, dkFor k moment characteristic distance matrixes, TkFor k-1
Pose to the k moment converts.Optimized using least square method and calculate Tk, according to TkUpdate point cloud position, carries out feature association again
And it updates distance d and calculates Tk, constantly cycle until d is less than given threshold or cycle-index reaches the upper limit.
(3) average value and comparison before optimization, n for seeking whole characteristic distances after the completion of optimizing are characterized a number.
If the average value of whole characteristic distances after optimization completion before optimization than reducing, then it is assumed that this adjustment completion；
Conversely, then to carry out adjustment optimization.
The failure of (4) adjustments is as caused by key frame similar features deficiency, this is needed to jacket key frame
It is expanded.
In map building imaging process, the point cloud of current key frame is mapped to after global coordinate system, it is possible to really
Determine position of the laser radar in the three-dimensional point cloud atlas created, in three-dimensional point cloud atlas, one is taken centered on this position
Cube in vivo all point clouds are with allowing current key frame and this partial dot cloud to be completed again by the progress of (2) the method is registering
Adjustment optimizes, wherein cubical size is related with the measurement range of laser radar, 16 line laser radar surveying model in this example
It encloses for 100m, therefore is all choosing the point cloud in the range of 100x100x30m in imaging point clouds.
As described above, the present invention utilizes the characteristic distributions of three-dimensional point cloud, key frame points cloud is chosen, then is chosen in key frame
Representational characteristic point, by completing adjustment in iterative calculation Optimized Matching position.This algorithm picks planar point and boundary point,
Universality is strong, can apply in various complicated mapping imaging and map building, while traversal search is most in registration process
Excellent solution moves selection all fixed points pair without operator player, improves operating efficiency.
Recited above is only the implementation for embodying the fast three-dimensional point cloud searching method for registering the present invention is based on cloud shape
Example.The present invention is not limited to above-described embodiments.The specification of the present invention is the model for not limiting claim for illustrating
It encloses.It will be apparent to those skilled in the art that can there are many replacements, improvement and variation.It is all to use equivalent substitution or wait
The technical solution that effect conversion is formed, all falls in the protection domain of the claims in the present invention.
Claims (2)
1. a kind of quick error compensation method in three-dimensional laser point cloud imaging, it is characterised in that：This method comprises the following steps：
Step (1) determines a fixed time interval, and key frame is chosen in whole point clouds of acquisition based on the interval
Adjustment is carried out, reduces computational complexity；
Step (2) extracts boundary point and planar point in key frame points cloud：For each point, the point is sought in its certain distance
The mould of the vector sum of interior all the points, then divided by laser range finder to the distance of the point, obtain the flatness i.e.Wherein i is the point calculated, and j is the consecutive points of i, and S is the point that the point of i both sides same number is formed
Collection, PiPjThe respectively coordinate of point i points j takes n point of wherein c values maximum as boundary point, the n of taking-up c values minimum (close to 0)
A point is as planar point；
After step (3) determines characteristic point, feature association is carried out to each characteristic point, feature association is exactly to seek this feature point to upper
The distance of the linked character of the point in one frame；
After step (4) feature association, each characteristic point finds one there are one the characteristic distance d on previous frame data
A pose converts Tk=[tx,ty,tz,rx,ry,rz], it is all 0 to allow all characteristic distances as far as possible, and the conversion of this pose is exactly to swash
Movement of the optical radar between this two frame data, i.e. f (Pk,Tk)=dk→ 0, PkFor k moment characteristic point matrix, dkIt is special for the k moment
Levy distance matrix, TkIt is converted for the pose at k-1 to k moment, is optimized using least square method and calculate Tk, according to TkUpdate point cloud position
It puts, carries out feature association again and update distance d to calculate Tk, constantly cycle until d is less than given threshold or cycle-index reaches
To the upper limit；
Step (5) reduces if the average value of whole characteristic distances after optimization completion is more preceding than optimization, then it is assumed that adjustment completion；Instead
It, then will carry out adjustment optimization：The point cloud of current key frame is mapped to after global coordinate system, it is possible to determine laser radar
Position in the three-dimensional point cloud atlas created, in three-dimensional point cloud atlas, taken centered on this position one cube it is in vivo
All point clouds with, allow current key frame and this partial dot cloud again by step (2), step (3), step (4) progress it is registering come
Optimize into adjustment, wherein cubical size is related with the measurement range of laser radar.
2. the quick error compensation method in a kind of three-dimensional laser point cloud imaging according to claim 1, it is characterised in that：It is described
Registration error method for solving is as follows in step (3)：
For step (3.1) if characteristic point is planar point, its linked character is a plane, it is necessary to find three in previous frame data
A point determines this plane：The point M nearest apart from characteristic point in previous keyframe, it is with scan line and distance M nearest with M
The point N nearest with M in the adjacent scanning lines of point L, M；
For step (3.2) if characteristic point is planar point, its linked character is a plane, it is necessary to find three in previous frame data
A point determines this plane：Point nearest with M in the adjacent scanning lines of the point M, M nearest apart from characteristic point in previous keyframe
N。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810021605.7A CN108090961B (en) | 2018-01-10 | 2018-01-10 | Rapid adjustment method in three-dimensional laser point cloud imaging |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810021605.7A CN108090961B (en) | 2018-01-10 | 2018-01-10 | Rapid adjustment method in three-dimensional laser point cloud imaging |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108090961A true CN108090961A (en) | 2018-05-29 |
| CN108090961B CN108090961B (en) | 2021-04-20 |
Family
ID=62181947
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810021605.7A Active CN108090961B (en) | 2018-01-10 | 2018-01-10 | Rapid adjustment method in three-dimensional laser point cloud imaging |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108090961B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112017202A (en) * | 2019-05-28 | 2020-12-01 | 杭州海康威视数字技术股份有限公司 | Point cloud labeling method, device and system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105447908A (en) * | 2015-12-04 | 2016-03-30 | 山东山大华天软件有限公司 | Dentition model generation method based on oral cavity scanning data and CBCT (Cone Beam Computed Tomography) data |
| CN107038717A (en) * | 2017-04-14 | 2017-08-11 | 东南大学 | A kind of method that 3D point cloud registration error is automatically analyzed based on three-dimensional grid |
-
2018
- 2018-01-10 CN CN201810021605.7A patent/CN108090961B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105447908A (en) * | 2015-12-04 | 2016-03-30 | 山东山大华天软件有限公司 | Dentition model generation method based on oral cavity scanning data and CBCT (Cone Beam Computed Tomography) data |
| CN107038717A (en) * | 2017-04-14 | 2017-08-11 | 东南大学 | A kind of method that 3D point cloud registration error is automatically analyzed based on three-dimensional grid |
Non-Patent Citations (8)
| Title |
|---|
| JIA CHEN 等: "3D shape modeling using a self-developed hand-held 3D laser scanner and an efficient HT-ICP point cloud registration algorithm", 《OPTICS &LASER TECHNOLOGY》 * |
| 严剑锋 等: "基于特征点提取和匹配的点云配准算法", 《测绘通报》 * |
| 张明明: "基于点云边界特征点的改进", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 张晓娟 等: "基于特征点和改进ICP的三维点云数据配准算法", 《传感器与微***》 * |
| 张步: "三维激光点云数据配准研究", 《万方数据企业知识服务平台》 * |
| 徐源强 等: "地面三维激光扫描的点云配准误差研究", 《大地测量与地球动力学》 * |
| 李仁忠 等: "基于ISS特征点结合改进ICP 的点云配准算法", 《激光与光电子学进展》 * |
| 胡少兴 等: "大型古文物真三维数字化方法", 《***仿真学报》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112017202A (en) * | 2019-05-28 | 2020-12-01 | 杭州海康威视数字技术股份有限公司 | Point cloud labeling method, device and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108090961B (en) | 2021-04-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111045017B (en) | Method for constructing transformer substation map of inspection robot by fusing laser and vision | |
| CN112505065B (en) | Method for detecting surface defects of large part by indoor unmanned aerial vehicle | |
| CN113436260B (en) | Mobile robot pose estimation method and system based on multi-sensor tight coupling | |
| CN104732518B (en) | A kind of PTAM improved methods based on intelligent robot terrain surface specifications | |
| CN108133458A (en) | A kind of method for automatically split-jointing based on target object spatial point cloud feature | |
| CN109509230A (en) | A kind of SLAM method applied to more camera lens combined type panorama cameras | |
| CN109993793B (en) | Visual positioning method and device | |
| CN111563442A (en) | Slam method and system for fusing point cloud and camera image data based on laser radar | |
| CN110146099B (en) | Synchronous positioning and map construction method based on deep learning | |
| CN109579825B (en) | Robot positioning system and method based on binocular vision and convolutional neural network | |
| CN111882612A (en) | Vehicle multi-scale positioning method based on three-dimensional laser detection lane line | |
| CN105809687A (en) | Monocular vision ranging method based on edge point information in image | |
| CN110136202A (en) | A kind of multi-targets recognition and localization method based on SSD and dual camera | |
| CN108680177B (en) | Synchronous positioning and map construction method and device based on rodent model | |
| CN111998862B (en) | BNN-based dense binocular SLAM method | |
| CN110223351B (en) | Depth camera positioning method based on convolutional neural network | |
| CN112484746B (en) | Monocular vision auxiliary laser radar odometer method based on ground plane | |
| CN113409459A (en) | Method, device and equipment for producing high-precision map and computer storage medium | |
| CN116449384A (en) | Radar inertial tight coupling positioning mapping method based on solid-state laser radar | |
| CN116468786A (en) | Semantic SLAM method based on point-line combination and oriented to dynamic environment | |
| CN114266821A (en) | Online positioning method and device, terminal equipment and storage medium | |
| CN108090961A (en) | A kind of quick error compensation method in three-dimensional laser point cloud imaging | |
| CN112767481A (en) | High-precision positioning and mapping method based on visual edge features | |
| CN113720323B (en) | Monocular vision inertial navigation SLAM method and device based on point-line feature fusion | |
| CN115290110A (en) | AR navigation method, system and computer readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |